kernel/linux.git/include, branch v3.10.61

mm: memcg: handle non-error OOM situations more gracefully

2014-11-21T17:22:56+00:00

commit 4942642080ea82d99ab5b653abb9a12b7ba31f4a upstream. Commit 3812c8c8f395 ("mm: memcg: do not trap chargers with full callstack on OOM") assumed that only a few places that can trigger a memcg OOM situation do not return VM_FAULT_OOM, like optional page cache readahead. But there are many more and it's impractical to annotate them all. First of all, we don't want to invoke the OOM killer when the failed allocation is gracefully handled, so defer the actual kill to the end of the fault handling as well. This simplifies the code quite a bit for added bonus. Second, since a failed allocation might not be the abrupt end of the fault, the memcg OOM handler needs to be re-entrant until the fault finishes for subsequent allocation attempts. If an allocation is attempted after the task already OOMed, allow it to bypass the limit so that it can quickly finish the fault and invoke the OOM killer. Reported-by: azurIt Signed-off-by: Johannes Weiner Cc: Michal Hocko Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Cong Wang Signed-off-by: Greg Kroah-Hartman

mm: memcg: do not trap chargers with full callstack on OOM

2014-11-21T17:22:56+00:00

commit 3812c8c8f3953921ef18544110dafc3505c1ac62 upstream. The memcg OOM handling is incredibly fragile and can deadlock. When a task fails to charge memory, it invokes the OOM killer and loops right there in the charge code until it succeeds. Comparably, any other task that enters the charge path at this point will go to a waitqueue right then and there and sleep until the OOM situation is resolved. The problem is that these tasks may hold filesystem locks and the mmap_sem; locks that the selected OOM victim may need to exit. For example, in one reported case, the task invoking the OOM killer was about to charge a page cache page during a write(), which holds the i_mutex. The OOM killer selected a task that was just entering truncate() and trying to acquire the i_mutex: OOM invoking task: mem_cgroup_handle_oom+0x241/0x3b0 mem_cgroup_cache_charge+0xbe/0xe0 add_to_page_cache_locked+0x4c/0x140 add_to_page_cache_lru+0x22/0x50 grab_cache_page_write_begin+0x8b/0xe0 ext3_write_begin+0x88/0x270 generic_file_buffered_write+0x116/0x290 __generic_file_aio_write+0x27c/0x480 generic_file_aio_write+0x76/0xf0 # takes ->i_mutex do_sync_write+0xea/0x130 vfs_write+0xf3/0x1f0 sys_write+0x51/0x90 system_call_fastpath+0x18/0x1d OOM kill victim: do_truncate+0x58/0xa0 # takes i_mutex do_last+0x250/0xa30 path_openat+0xd7/0x440 do_filp_open+0x49/0xa0 do_sys_open+0x106/0x240 sys_open+0x20/0x30 system_call_fastpath+0x18/0x1d The OOM handling task will retry the charge indefinitely while the OOM killed task is not releasing any resources. A similar scenario can happen when the kernel OOM killer for a memcg is disabled and a userspace task is in charge of resolving OOM situations. In this case, ALL tasks that enter the OOM path will be made to sleep on the OOM waitqueue and wait for userspace to free resources or increase the group's limit. But a userspace OOM handler is prone to deadlock itself on the locks held by the waiting tasks. For example one of the sleeping tasks may be stuck in a brk() call with the mmap_sem held for writing but the userspace handler, in order to pick an optimal victim, may need to read files from /proc/, which tries to acquire the same mmap_sem for reading and deadlocks. This patch changes the way tasks behave after detecting a memcg OOM and makes sure nobody loops or sleeps with locks held: 1. When OOMing in a user fault, invoke the OOM killer and restart the fault instead of looping on the charge attempt. This way, the OOM victim can not get stuck on locks the looping task may hold. 2. When OOMing in a user fault but somebody else is handling it (either the kernel OOM killer or a userspace handler), don't go to sleep in the charge context. Instead, remember the OOMing memcg in the task struct and then fully unwind the page fault stack with -ENOMEM. pagefault_out_of_memory() will then call back into the memcg code to check if the -ENOMEM came from the memcg, and then either put the task to sleep on the memcg's OOM waitqueue or just restart the fault. The OOM victim can no longer get stuck on any lock a sleeping task may hold. Debugged by Michal Hocko. Signed-off-by: Johannes Weiner Reported-by: azurIt Acked-by: Michal Hocko Cc: David Rientjes Cc: KAMEZAWA Hiroyuki Cc: KOSAKI Motohiro Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Cong Wang Signed-off-by: Greg Kroah-Hartman

mm: memcg: enable memcg OOM killer only for user faults

2014-11-21T17:22:56+00:00

commit 519e52473ebe9db5cdef44670d5a97f1fd53d721 upstream. System calls and kernel faults (uaccess, gup) can handle an out of memory situation gracefully and just return -ENOMEM. Enable the memcg OOM killer only for user faults, where it's really the only option available. Signed-off-by: Johannes Weiner Acked-by: Michal Hocko Cc: David Rientjes Cc: KAMEZAWA Hiroyuki Cc: azurIt Cc: KOSAKI Motohiro Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds Signed-off-by: Cong Wang Signed-off-by: Greg Kroah-Hartman

arch: mm: pass userspace fault flag to generic fault handler

2014-11-21T17:22:56+00:00

commit 759496ba6407c6994d6a5ce3a5e74937d7816208 upstream. Unlike global OOM handling, memory cgroup code will invoke the OOM killer in any OOM situation because it has no way of telling faults occuring in kernel context - which could be handled more gracefully - from user-triggered faults. Pass a flag that identifies faults originating in user space from the architecture-specific fault handlers to generic code so that memcg OOM handling can be improved. Signed-off-by: Johannes Weiner Reviewed-by: Michal Hocko Cc: David Rientjes Cc: KAMEZAWA Hiroyuki Cc: azurIt Cc: KOSAKI Motohiro Signed-off-by: Andrew Morton Signed-off-by: Cong Wang Signed-off-by: Linus Torvalds Signed-off-by: Greg Kroah-Hartman

net: sctp: fix skb_over_panic when receiving malformed ASCONF chunks

2014-11-21T17:22:55+00:00

commit 9de7922bc709eee2f609cd01d98aaedc4cf5ea74 upstream. Commit 6f4c618ddb0 ("SCTP : Add paramters validity check for ASCONF chunk") added basic verification of ASCONF chunks, however, it is still possible to remotely crash a server by sending a special crafted ASCONF chunk, even up to pre 2.6.12 kernels: skb_over_panic: text:ffffffffa01ea1c3 len:31056 put:30768 head:ffff88011bd81800 data:ffff88011bd81800 tail:0x7950 end:0x440 dev: ------------[ cut here ]------------ kernel BUG at net/core/skbuff.c:129! [...] Call Trace: [] skb_put+0x5c/0x70 [] sctp_addto_chunk+0x63/0xd0 [sctp] [] sctp_process_asconf+0x1af/0x540 [sctp] [] ? _read_unlock_bh+0x15/0x20 [] sctp_sf_do_asconf+0x168/0x240 [sctp] [] sctp_do_sm+0x71/0x1210 [sctp] [] ? fib_rules_lookup+0xad/0xf0 [] ? sctp_cmp_addr_exact+0x32/0x40 [sctp] [] sctp_assoc_bh_rcv+0xd3/0x180 [sctp] [] sctp_inq_push+0x56/0x80 [sctp] [] sctp_rcv+0x982/0xa10 [sctp] [] ? ipt_local_in_hook+0x23/0x28 [iptable_filter] [] ? nf_iterate+0x69/0xb0 [] ? ip_local_deliver_finish+0x0/0x2d0 [] ? nf_hook_slow+0x76/0x120 [] ? ip_local_deliver_finish+0x0/0x2d0 [] ip_local_deliver_finish+0xdd/0x2d0 [] ip_local_deliver+0x98/0xa0 [] ip_rcv_finish+0x12d/0x440 [] ip_rcv+0x275/0x350 [] __netif_receive_skb+0x4ab/0x750 [] netif_receive_skb+0x58/0x60 This can be triggered e.g., through a simple scripted nmap connection scan injecting the chunk after the handshake, for example, ... -------------- INIT[ASCONF; ASCONF_ACK] -------------> <----------- INIT-ACK[ASCONF; ASCONF_ACK] ------------ -------------------- COOKIE-ECHO --------------------> <-------------------- COOKIE-ACK --------------------- ------------------ ASCONF; UNKNOWN ------------------> ... where ASCONF chunk of length 280 contains 2 parameters ... 1) Add IP address parameter (param length: 16) 2) Add/del IP address parameter (param length: 255) ... followed by an UNKNOWN chunk of e.g. 4 bytes. Here, the Address Parameter in the ASCONF chunk is even missing, too. This is just an example and similarly-crafted ASCONF chunks could be used just as well. The ASCONF chunk passes through sctp_verify_asconf() as all parameters passed sanity checks, and after walking, we ended up successfully at the chunk end boundary, and thus may invoke sctp_process_asconf(). Parameter walking is done with WORD_ROUND() to take padding into account. In sctp_process_asconf()'s TLV processing, we may fail in sctp_process_asconf_param() e.g., due to removal of the IP address that is also the source address of the packet containing the ASCONF chunk, and thus we need to add all TLVs after the failure to our ASCONF response to remote via helper function sctp_add_asconf_response(), which basically invokes a sctp_addto_chunk() adding the error parameters to the given skb. When walking to the next parameter this time, we proceed with ... length = ntohs(asconf_param->param_hdr.length); asconf_param = (void *)asconf_param + length; ... instead of the WORD_ROUND()'ed length, thus resulting here in an off-by-one that leads to reading the follow-up garbage parameter length of 12336, and thus throwing an skb_over_panic for the reply when trying to sctp_addto_chunk() next time, which implicitly calls the skb_put() with that length. Fix it by using sctp_walk_params() [ which is also used in INIT parameter processing ] macro in the verification *and* in ASCONF processing: it will make sure we don't spill over, that we walk parameters WORD_ROUND()'ed. Moreover, we're being more defensive and guard against unknown parameter types and missized addresses. Joint work with Vlad Yasevich. Fixes: b896b82be4ae ("[SCTP] ADDIP: Support for processing incoming ASCONF_ACK chunks.") Signed-off-by: Daniel Borkmann Signed-off-by: Vlad Yasevich Acked-by: Neil Horman Signed-off-by: David S. Miller Cc: Josh Boyer Signed-off-by: Greg Kroah-Hartman

net: sctp: fix panic on duplicate ASCONF chunks

2014-11-21T17:22:55+00:00

commit b69040d8e39f20d5215a03502a8e8b4c6ab78395 upstream. When receiving a e.g. semi-good formed connection scan in the form of ... -------------- INIT[ASCONF; ASCONF_ACK] -------------> <----------- INIT-ACK[ASCONF; ASCONF_ACK] ------------ -------------------- COOKIE-ECHO --------------------> <-------------------- COOKIE-ACK --------------------- ---------------- ASCONF_a; ASCONF_b -----------------> ... where ASCONF_a equals ASCONF_b chunk (at least both serials need to be equal), we panic an SCTP server! The problem is that good-formed ASCONF chunks that we reply with ASCONF_ACK chunks are cached per serial. Thus, when we receive a same ASCONF chunk twice (e.g. through a lost ASCONF_ACK), we do not need to process them again on the server side (that was the idea, also proposed in the RFC). Instead, we know it was cached and we just resend the cached chunk instead. So far, so good. Where things get nasty is in SCTP's side effect interpreter, that is, sctp_cmd_interpreter(): While incoming ASCONF_a (chunk = event_arg) is being marked !end_of_packet and !singleton, and we have an association context, we do not flush the outqueue the first time after processing the ASCONF_ACK singleton chunk via SCTP_CMD_REPLY. Instead, we keep it queued up, although we set local_cork to 1. Commit 2e3216cd54b1 changed the precedence, so that as long as we get bundled, incoming chunks we try possible bundling on outgoing queue as well. Before this commit, we would just flush the output queue. Now, while ASCONF_a's ASCONF_ACK sits in the corked outq, we continue to process the same ASCONF_b chunk from the packet. As we have cached the previous ASCONF_ACK, we find it, grab it and do another SCTP_CMD_REPLY command on it. So, effectively, we rip the chunk->list pointers and requeue the same ASCONF_ACK chunk another time. Since we process ASCONF_b, it's correctly marked with end_of_packet and we enforce an uncork, and thus flush, thus crashing the kernel. Fix it by testing if the ASCONF_ACK is currently pending and if that is the case, do not requeue it. When flushing the output queue we may relink the chunk for preparing an outgoing packet, but eventually unlink it when it's copied into the skb right before transmission. Joint work with Vlad Yasevich. Fixes: 2e3216cd54b1 ("sctp: Follow security requirement of responding with 1 packet") Signed-off-by: Daniel Borkmann Signed-off-by: Vlad Yasevich Signed-off-by: David S. Miller Cc: Josh Boyer Signed-off-by: Greg Kroah-Hartman

netfilter: xt_bpf: add mising opaque struct sk_filter definition

2014-11-21T17:22:54+00:00

commit e10038a8ec06ac819b7552bb67aaa6d2d6f850c1 upstream. This structure is not exposed to userspace, so fix this by defining struct sk_filter; so we skip the casting in kernelspace. This is safe since userspace has no way to lurk with that internal pointer. Fixes: e6f30c7 ("netfilter: x_tables: add xt_bpf match") Signed-off-by: Pablo Neira Ayuso Acked-by: Willem de Bruijn Signed-off-by: David S. Miller Signed-off-by: Greg Kroah-Hartman

clocksource: Remove "weak" from clocksource_default_clock() declaration

2014-11-21T17:22:54+00:00

commit 96a2adbc6f501996418da9f7afe39bf0e4d006a9 upstream. kernel/time/jiffies.c provides a default clocksource_default_clock() definition explicitly marked "weak". arch/s390 provides its own definition intended to override the default, but the "weak" attribute on the declaration applied to the s390 definition as well, so the linker chose one based on link order (see 10629d711ed7 ("PCI: Remove __weak annotation from pcibios_get_phb_of_node decl")). Remove the "weak" attribute from the clocksource_default_clock() declaration so we always prefer a non-weak definition over the weak one, independent of link order. Fixes: f1b82746c1e9 ("clocksource: Cleanup clocksource selection") Signed-off-by: Bjorn Helgaas Acked-by: John Stultz Acked-by: Ingo Molnar CC: Daniel Lezcano CC: Martin Schwidefsky Signed-off-by: Greg Kroah-Hartman

kgdb: Remove "weak" from kgdb_arch_pc() declaration

2014-11-21T17:22:54+00:00

commit 107bcc6d566cb40184068d888637f9aefe6252dd upstream. kernel/debug/debug_core.c provides a default kgdb_arch_pc() definition explicitly marked "weak". Several architectures provide their own definitions intended to override the default, but the "weak" attribute on the declaration applied to the arch definitions as well, so the linker chose one based on link order (see 10629d711ed7 ("PCI: Remove __weak annotation from pcibios_get_phb_of_node decl")). Remove the "weak" attribute from the declaration so we always prefer a non-weak definition over the weak one, independent of link order. Fixes: 688b744d8bc8 ("kgdb: fix signedness mixmatches, add statics, add declaration to header") Tested-by: Vineet Gupta # for ARC build Signed-off-by: Bjorn Helgaas Reviewed-by: Harvey Harrison Signed-off-by: Greg Kroah-Hartman

nfs: fix pnfs direct write memory leak

2014-11-21T17:22:53+00:00

commit 8c393f9a721c30a030049a680e1bf896669bb279 upstream. For pNFS direct writes, layout driver may dynamically allocate ds_cinfo.buckets. So we need to take care to free them when freeing dreq. Ideally this needs to be done inside layout driver where ds_cinfo.buckets are allocated. But buckets are attached to dreq and reused across LD IO iterations. So I feel it's OK to free them in the generic layer. Signed-off-by: Peng Tao Signed-off-by: Trond Myklebust Signed-off-by: Greg Kroah-Hartman